Using longitudinal data to understand the impact of car sharing on car ownership: An analysis based on the German Mobility Panel (MOP)

Car sharing is a shared mobility service that allows renting a car for a short time or distance without the burden of complete ownership, such as maintenance, insurance, and repair responsibilities. Car sharing, as one of the newest additions to transportation modes, has the potential to attract a significant number of drivers. In recent decades, academia and government have increasingly paid attention to car sharing and how it affects the urban transportation system. One of the most important indicators for evaluating car sharing benefits is its impact on household car ownership. Although the effects of car sharing on household vehicle holdings have been extensively studied, there is a lack of dynamic analysis through a panel survey, despite the fact that the effects of car sharing on car ownership is a dynamic process that takes place over time. In this study, we perform the analysis using data from the 2012-2020 German Mobility Panel (MOP), an unbalanced and rotating panel survey conducted annually since 1994. All the German-speaking households living in Germany can voluntarily participate. The MOP sample size is provided in two household and person levels, with 1173 households and 1913 individuals in 2012. It then recorded 80% growth, reaching 3461 individuals in 2020. Noticeably, the sampling method is rotating, and three years of information are the maximum available data for each interviewee. This thesis aims at understanding the impact of car sharing on car ownership, and it investigates in light of early studies' limitations, such as casualty effects, self-selection, and recall biases. This analysis uses longitudinal panel data to prevent recall bias. Moreover, propensity-score-based matching is used to help control self-selection bias due to differences in observed socio-demographic characteristics between respondents. The treatment and control groups are identified to isolate car sharing membership effect and establish causal relationships. It is noted that control groups contain never car sharing members. We select the nearest neighbor matching method with a ratio of five for this thesis. Both control and treated units are estimated using logistic regression using R software. We identify the treated group for each wave of the MOP survey, finding that 115 unique car sharing members with at least one year of information fit the dynamic analysis purpose. Having paid attention to the dynamic behavior, among 5 members showing an increasing trend in their car holdings, 3 bought a new car in the same year of unsubscription to a car sharing program. In addition, 5 out of 6 members sold a car when they subscribed to a car sharing scheme. Afterward, we found 676 unique matched IDs by performing the matching method. With car ownership patterns of all the treated and control units, it is then possible to perform a matched-pair comparison. In both groups, most units do not change their car ownership. 4.7% of the control groups purchased a new car, but 4.3% of the treated ones. Oppositely, more car sharing users foregone a vehicle than the controls, amounting to 5.2% and 4%. Although such results are based on a small number of observations related to car ownership changes of car sharing members, we finally project them to the whole universe (German car drivers) to show that car sharing subscription and unsubscription versus car ownership changes are not represented by symmetric patterns, therefore cross-sectional data fail in understanding the real impact of car sharing on car ownership.